North Atlantic warming: patterns of long-term trend and multidecadal variability
- 1.8k Downloads
Climate fluctuations in the North Atlantic Ocean have wide-spread implications for Europe, Africa, and the Americas. This study assesses the relative contribution of the long-term trend and variability of North Atlantic warming using EOF analysis of deep-ocean and near-surface observations. Our analysis demonstrates that the recent warming over the North Atlantic is linked to both long-term (including anthropogenic and natural) climate change and multidecadal variability (MDV, ~50–80 years). Our results suggest a general warming trend of 0.031 ± 0.006°C/decade in the upper 2,000 m North Atlantic over the last 80 years of the twentieth century, although during this time there are periods in which short-term trends were strongly amplified by MDV. For example, MDV accounts for ~60% of North Atlantic warming since 1970. The single-sign basin-scale pattern of MDV with prolonged periods of warming (cooling) in the upper ocean layer and opposite tendency in the lower layer is evident from observations. This pattern is associated with a slowdown (enhancement) of the North Atlantic thermohaline overturning circulation during negative (positive) MDV phases. In contrast, the long-term trend exhibits warming in tropical and mid-latitude North Atlantic and a pattern of cooling in regions associated with major northward heat transports, consistent with a slowdown of the North Atlantic circulation as evident from observations and confirmed by selected modeling results. This localized cooling has been masked in recent decades by warming during the positive phase of MDV. Finally, since the North Atlantic Ocean plays a crucial role in establishing and regulating the global thermohaline circulation, the multidecadal fluctuations discussed here should be considered when assessing long-term climate change and variability, both in the North Atlantic and at global scales.
KeywordsNorth Atlantic Multidecadal variability Climate change
Substantial North Atlantic climate changes have occurred over recent decades (e.g. Levitus et al. 2000). Despite the fact that these changes have contributed to observed warming of the Northern Hemisphere, including rapid warming over Europe and high-latitude regions, and to changes in terrestrial and marine ecosystems, an understanding of the governing mechanisms and their attribution either to human-induced (anthropogenic) climate change or to natural variability has not been well established (IPCC 2007). This uncertainty has sparked debate about the relative roles of multidecadal variability (MDV) versus anthropogenic forcing in the observed warming in the tropical Atlantic, which is closely related to the Atlantic hurricane activity (Mann and Emanuel 2006). Much of this debate centers around how to define the strength of the multidecadal variations which occur on a timescale of 50–80 years and have large amplitude variations in the North Atlantic (Schlesinger and Ramankutty 1994).
These low-frequency fluctuations are evident in various instrumental and proxy records from the Northern Hemisphere (see Delworth and Mann 2000 for references therein) but much is not understood about MDV. A study of the proxy and long-term instrumental records by Stocker and Mysak (1992) emphasized that these low frequency variations, though global in extent, are most pronounced in the Atlantic Ocean. Folland et al. (1986) reached the same conclusion from an analysis of global sea surface temperatures (SST). A time series of area-averaged North Atlantic SST with the local trend removed defines the so called Atlantic multidecadal oscillation (AMO, Enfield et al. (2001)). There are, however arguments that the observed SST changes are not consistent with the linear North Atlantic trend used to define the AMO. A revised AMO index was proposed in which the global SST trend was removed instead. Defined this way, the revised AMO accounts for only 0–0.1°C of the recent North Atlantic SST anomaly, several times less than what would be accounted for by the “standard” AMO (Trenberth and Shea 2006). Parker et al. (2007) applied empirical orthogonal function (EOF) analysis to a global SST dataset and argued that the AMO appears as the third EOF with a weak trend similar to that from Trenberth and Shea (2006).
The relatively short instrumental records of the AMO are augmented with paleoclimate data to provide evidence of a multidecadal signal over several centuries (e.g. Gray et al. 2004; Divine and Dick 2006; Fritzsche et al. 2005). Part of the difficulty in identifying the low-frequency fluctuations and understanding mechanisms behind the MDV is due to it’s evolving spectrum, and the changing relationship with between SAT/SST (SAT, surface air temperature) and the large-scale atmospheric forcing like the North Atlantic oscillation (NAO, defined as the north–south-oriented dipole in sea-level pressure (SLP) over the Atlantic) (Polyakova et al. 2006). For example, in contrast to the warming of the 1990s, the 1930s warm period in the Arctic did not coincide with a strongly positive phase of the NAO (Overland et al. 2004), as would be expected based on the NAO paradigm of greater heat transport into the Arctic when the NAO is positive. This led to the conjecture that the mechanism of the warming of the 1930s was associated with local air–sea–ice interactions (Bengtsson et al. 2004) and the recent warming was due to a different mechanism. Atmospheric heat-transport mechanisms which are important for high-latitude heat budget but at the same time unrelated to the positive surface albedo feedback were described in (Alexeev 2003; Alexeev et al. 2005; Langen and Alexeev 2007). An extensive analysis of Arctic and North Atlantic atmosphere, ocean, and ice observations demonstrates that there are many similarities between these two warm periods, suggesting that both periods are associated with related mechanisms (e.g. Polyakov et al. 2008).
Nevertheless, numerous studies (e.g. Bjerkness 1964; Deser and Blackmon 1993; Kushnir 1994; Dickson et al. 1996, 2002; Timmermann et al. 1998; Curry et al. 1998, 2003; Häkkinen 1999; Curry and McCartney 2001; Visbeck et al. 2002; Peterson et al. 2006) point to the important role of the NAO in climate variability in the North Atlantic Ocean, making it critical to understand why the NAO paradigm does not always operate.
While the physical mechanisms for generating MDV may differ from model to model (e.g. Latif 1998), there is a consensus that long-term changes in the thermohaline (e.g. density-driven) circulation (THC) play a crucial role in establishing spatial and temporal SST patterns (e.g. Delworth et al. 1993; Timmermann et al. 1998; Häkkinen 1999; Delworth and Greatbatch 2000; Eden and Jung 2001; Barnett et al. 2005; Knight et al. 2005; Hawkins and Sutton 2007; Zhang et al. 2007). The North Atlantic THC is associated with the meridional overturning circulation (MOC), a northward flow of warm, low-density surface waters balanced by a commensurate southward flow of cold, high-density waters at depth. There is compelling evidence to support the notion that phases of MDV expressed by the AMO and the strength of the MOC are interrelated. For example, Parker et al. (2007) analyzed modeling results and demonstrated that a weak (strong) MOC is associated with a cooler (warmer) North Atlantic. Parker et al. also demonstrated that models encounter substantial problems in simulating MDV of the NAO.
Thus, it is imperative to understand long-term fluctuations in the North Atlantic by linking changes at the ocean surface and in the ocean interior with possible changes in sea surface height, strength of the oceanic circulation and air–ocean interactions. This study assesses the relative contribution of the long-term trend and variability to warming in the North Atlantic by evaluating these processes without any a priori assumptions about their shape. In this paper, we use the term “long-term climate change” without distinguishing between anthropogenic climate change and additional (longer than multidecadal) low-frequency natural variations that are not resolved by the observational records in accordance with the IPCC (2007).
2 Data and methods
The major emphasis of this study comprises an analysis of observations but is augmented with a model analysis that focuses on the climate dynamics of the long-term trend. For this purpose, we used twentieth century (1900–1999, 20C3M) model runs from four fully coupled global climate models: BCCR-BCM2.0, CCSM3, GFDL-CM2.0, and UKMO-HadCM3. These models were among a suite used for the latest Intergovernmental Panel on Climate Change (IPCC) Fourth Assessment Report (2007). These four models were selected because their simulation of north–south long-term contrast of the North Atlantic SST is in reasonably good agreement with observations. Detailed information about the models and their simulation design can be found at http://www-pcmdi.llnl.gov/ipcc/model_documentation/ipcc_model_documentation.php and the embedded modeling centers’ website links. The climate of the twentieth century is forced with observed emissions which includes greenhouse gas concentrations and, in some cases, aerosol and variable solar forcing. In order to avoid potential impacts from model spin-up, all the models were initialized at various antecedent times early in the mid- or late-nineteenth century. When processing model data, we followed the analysis procedure used for the observational data as closely as possible: model data were averaged within the same zonal belts and interpolated vertically onto the vertical grid used for analysis of the observational data. This approach facilitates the comparison of observed and simulated EOFs of zonal mean temperatures.
The net radiative forcing used to interpret our results was obtained from http://data.giss.nasa.gov/modelforce/ by computing a cumulative time series with the mean removed based on data prior 1900 (in the text, we will refer to this parameter as “radiative forcing”). Maximenko (IPRC) and Niiler (SIO) provided ocean dynamic topography data. ICOADS Sensible Heat Flux data (ICOADS 2-degree Enhanced Release 2.3) were provided by the NOAA/OAR/ESRL PSD, Boulder, Colorado, USA from http://www.cdc.noaa.gov/. SLP (5 × resolution, 15–90 N) was provided by NCAR based on the methods of Trenberth and Paolino (1980) from www.cgd.ucar.edu/cas/guide/Data/trenpaol.html. SST data (5 × 5 Kaplan extended SST v.2) were obtained from NCAR at www.cgd.ucar.edu/cas/guide/Data/kaplan_sst.html (Kaplan et al. 1998). Regression analysis of SLP, SST, and heat fluxes on the PCs was performed in order to illustrate the spatiotemporal behavior of these parameters.
3 Separating leading modes of North Atlantic variability
Trends (°C/decade) of principal components (PCs) of the EOF analysis of zonal average North Atlantic water temperature
0.031 ± 0.006
0.045 ± 0.007
0.043 ± 0.011
0.060 ± 0.036
0.006 ± 0.007
0.008 ± 0.011
0.014 ± 0.013
0.081 ± 0.019
4 Features of North Atlantic multidecadal variability
The striking resemblance between MDV of the North Atlantic, whether expressed by PC2, SST, or upper 300 m water temperature and salinity anomalies (Fig. 3), supports the hypothesis that “universal” mechanisms govern North Atlantic large-scale low-frequency variations. Numerous studies point to the importance of the NAO in forcing low-frequency variations in the North Atlantic Ocean. Our analysis suggests that some observed features of North Atlantic MDV may be associated with the NAO. For example, the maximum statistically significant correlation of R = 0.48 between the NAO and PC2 is found with an 11-year lag, which is consistent with earlier findings of a delayed oceanic response to atmospheric forcing (e.g. Eden and Willebrand 2001). However, the SLP anomalies obtained by regressing lagged SLP on PC2 (Fig. 5) agree better with the East Atlantic pattern (not shown), one of the leading EOF modes of the SLP (Barnston and Livezey 1987). Regression analysis of SST on PC2 reveals a single-sign basin-scale anomaly pattern characteristic of multi-decadal fluctuations in the North Atlantic SST (Fig. 5) consistent with earlier findings (e.g. Bjerkness 1964; Kushnir 1994; Delworth and Greatbatch 2000; and Eden and Willebrand 2001; see also Polyakov et al. (2005) and Visbeck et al. (2002) for further discussion and references therein). Multidecadal fluctuations in the zonal mean water temperature as expressed by the EOF2 (Fig. 2) are dominated by warming (cooling) in the upper 300 m layer and cooling (warming) in the 1,000–3,000 m layer during the positive (negative) MDV phase. Note that the deeper ocean displays anomalies that are approximately 40% of the upper ocean anomalies. This is generally consistent with changes in SST (Fig. 5) and supported by earlier studies (Polyakov et al. 2005; Zhang 2007). For example, using observational data, Polyakov et al. (2005) demonstrated that temperature and salinity from the 0–300 to 1,000–3,000 m layers vary in opposition: prolonged periods of cooling and freshening (warming and salinification) in one layer are generally associated with opposite tendencies in the other layer. This pattern may be associated with a change in the strength of MOC.
The response of large-scale oceanic circulation to atmospheric forcing is further evaluated using the theory of Marshall et al. (2001a, b). They showed that a circulation anomaly called the ‘intergyre’ gyre (its approximate position is shown by the box in Fig. 5) is driven by meridional shifts in the zero wind curl which is climatologically located between the subpolar and subtropical gyres. According to Marshall et al. (2001a, b), the oscillatory behavior of the intergyre gyre is governed by north–south heat transports by anomalous currents, balanced by damping of the SST anomalies via air–sea interactions. Our analysis provides further evidence supporting the important role played by the intergyre gyre in establishing and regulating multidecadal temperature variations of the North Atlantic (Fig. 5). For example, MDV of the zonal average temperatures expressed by PC2 and wind curl anomalies computed over the intergyre gyre region (Fig. 3) are negatively correlated (R = −0.44) at an 8-year lag. This is consistent with a delayed oceanic response to the atmospheric forcing found in modeling (Eden and Willebrand 2001) and theoretical (Marshall et al. 2001a, b) studies. During prolonged phases of high (low) wind vorticity there is anomalous upwelling (downwelling) centered at ~45N concurrent with lower (higher) SSTs and decreased (increased) surface heat fluxes out of the ocean (Fig. 5). This pattern is consistent with the ocean response to the NAO simulated by GCMs (e.g. Eden and Willebrand 2001; Vellinga and Wu 2004) and is confirmed by the statistically significant minimum in the regression pattern of SST on wind vorticity index lagged by 8 years (not shown). We find that the change of density structure during cold (warm) phases of MDV suppresses (enhances) the intergyre gyre as seen from increased (decreased) SSH slopes when the circulation is enhanced (suppressed) (Fig. 7). An important implication for Arctic–North Atlantic interactions is that the intergyre dynamics introduces much shorter timescales than those imposed by the thermohaline circulation (planetary-scale conveyor) (Marshall et al. 2001a, b).
5 Long-term trend and multidecadal variability from four IPCC models
Analysis of model data is necessary for developing an understanding of causality of climate processes since observations result from all processes and feedbacks in nature while models represent only a part of the fully-coupled system and often not all needed variables are observed. In this section, model data are used to gain insight about governing forces driving observed long-term changes hypothesized from the analysis of Sects. 3–4. Specifically, we use modeling results to gain a better understanding of the observed long-term cooling of the northern North Atlantic as expressed by EOF1. Multi-model ensembles have been used by Kravtsov and Spannagle (2008) to estimate the contribution of radiative forcing to twentieth century trends. The readily available multi-model IPCC twentieth century scenario data archives are the ideal tool for such an exercise. We focus on four general circulation models that participated in the recent IPCC Report (2007) and repeat the same analysis on model data as was done for the observations in the previous sections. These four models were selected because their simulation of north (cool)–south (warm) long-term contrast of the North Atlantic SST is in reasonably good agreement with observations. The model results are compared with the observed analysis to evaluate the models as well as explore possible mechanisms. Note that the objectives of this analysis are somewhat restricted - we do not attempt to explore physical mechanisms in-depth which would require performing a suite of model sensitivity simulations but rather use available simulations to determine what types of responses are possible in the models.
The IPCC models display varying success at simulating features of MDV. The second EOFs with the corresponding PCs of simulated zonal mean water temperature which, according to our observational analysis is associated with the multidecadal mode of variability are shown in Fig. 8 (right). The twentieth century model runs are dominated by the long-term trend as expressed by greater variance explained by EOF1s; as a result the relative role of multidecadal fluctuations is proportionally less in the modeled data than in the observations. The modeled EOF2 patterns bear certain similarities (Fig. 8, right) to the observations. For example, all models show that during the positive phase the northern North Atlantic region (>50 N) is warmer down to 1,000–1,500 m and deeper depending on the model. The simulated cold anomaly associated with the intergyre gyre at ~40–45 N occupies a substantial portion of the ocean interior while the simulated variability in the tropical North Atlantic is not as consistent among the models. The HadCM3 run seems to be the most successful in simulating the observed pattern of multidecadal fluctuations showing opposing anomalies in the upper and lower ocean (compare Figs. 2, 8) and a single-sign basin-wide spatial pattern of MDV in each layer (compare Figs. 5, 9). The BCCR-BCM2.0 simulation of the THC driving these changes is shown in Fig. 11. The EOF2 pattern from BCCR-BCM2.0 compares favorably with our observation-based findings of suppressed (enhanced) circulation in the tropical North Atlantic (intergyre gyre) (with less success in reproducing the subpolar basin) during the positive MDV phase. However, PC2 for BCCR-BCM2.0 MOC appears noisy and contains decadal-scale variations which mask the multidecadal signal. Note that a similar EOF pattern for MDV was obtained by Eden and Willebrand (2001) as a response of the ocean to changes in atmospheric circulation and by Vellinga and Wu (2004) in a HadCM3 control run as a signature of internal oceanic THC. It has been stressed by Osborn (2004), and our analysis confirms this conclusion, that the observed NAO pattern which is enhanced (suppressed) during the positive (negative) phase of MDV is not well reproduced by all GCMs. For example, the CCSM3.0 simulation shows quite a different SLP pattern (Fig. 12) compared with observations (Fig. 5). Two modeled SLP distributions (HadCM3 and BCCR-BCM2.0) are similar to each other, but look less zonal compared with the observed pattern.
The reasons for inconsistencies between models and observations may be different (see Sect. 6 in depth in Parker et al. 2007) and one of the purposes of using several simulations in this study was to demonstrate this diversity. In this study we have not explored important physical mechanisms like convective ventilation (e.g. Hawkins and Sutton 2007), air–sea interactions (e.g. Timmermann et al. 1998; Bhatt et al. 1998) or internal oceanic THC (e.g. Vellinga and Wu 2004). However, the modeling results provide support for our observation-based conclusions namely that the climate change-related pattern of regional cooling is associated with major northward heat transports and is consistent with a slowdown of the North Atlantic circulation.
6 Discussion and conclusions
6.1 Long-term trend
An analysis of observational ocean temperatures complemented with modeling results is used to assess the relative contributions of the long-term trend and large-amplitude multidecadal fluctuations to warming in the North Atlantic. The modal structure of North Atlantic variability derived from observations and modeling may be summarized as follows. The leading mode of oceanic variability captures the long-term non-linear trend which displays an accelerated increase in recent decades and we speculate that it may be related to enhanced radiative forcing. It is still unclear why the zero-lag correlation between PC1 of the zonal mean water temperature and the net radiative forcing is maximum (Fig. 3) with correlations decreasing rapidly with increasing lag. Assuming this is true then one interpretation is that the fast oceanic response to radiative forcing may be due to the strong impact that convective processes have on the formation rate of the North Atlantic Intermediate Water in the Labrador Sea (Dickson et al. 1996, 2002; Curry and McCartney 2001; Curry et al. 2003). Convectively-driven Labrador Sea anomalies spread across the northern North Atlantic surprisingly quickly (Sy et al. 1997). Modeling results support the important role of Labrador Sea convection in shaping North Atlantic multidecadal fluctuations (e.g. Jungclaus et al. 2005; Hawkins and Sutton 2007). Another possibility is that the seemingly fast deep-ocean response to radiative forcing may be linked to the barotropic mode excited by the radiative forcing via SSH modulations. However, additional research is necessary to explain this further.
The spatial structure of the leading mode may be expressed in terms of a large-scale horizontal gyre-like circulation. One of the most intriguing features of the long-term warming trend is the presence of anomalously cold water located in the regions of major North Atlantic surface heat transports, including the Gulf Stream—North Atlantic Current system and their poleward continuation, the Norwegian Current, consistent with a slowdown of the North Atlantic circulation. The pattern of North Atlantic cooling north of the mean zero wind stress curl line and warming southward (Fig. 5) also suggests that the signal may partially result from an increased wind stress curl driving increased upwelling of cold waters at high latitudes and more pumping of warm waters at low latitudes. Anomalous advection of cold northerly air masses due to changes in atmospheric circulation may play a role as well.
6.2 Multidecadal variability
Temperature and salinity in the upper and lower layers vary in opposition, consistent with the notion of THC, with prolonged periods of basin-wide single-sign warming and salinification (cooling and freshening) in one layer associated with the opposite tendencies in the other layer (Visbeck et al. 2002; Polyakov et al. 2005; Zhang et al. 2007, Figs. 2, 5 from this study).
Convergence (divergence) of winds (Fig. 3) drives an oceanic downwelling (upwelling) centered at ~45 N (intergyre gyre, Eden and Willebrand 2001; Marshall et al. 2001a, b) and is characterized by anomalously high (low) SSTs and enhanced (reduced) surface heat fluxes from the ocean to the atmosphere during the positive (negative) phase of MDV (e.g. Marshall et al. 2001a, b, see also Fig. 5).
Changes in the density structure between the warm and cold phases of MDV act in an opposing way on the ocean circulation by enhancing (suppressing) the intergyre gyre centered around ~45 N while at the same time weakening (strengthening) the northeastern flow in the northern (~52–65 N) North Atlantic and in the tropics (<30 N) (Häkkinen and Rhines 2004; Bryden et al. 2005; Fig. 6).
These changes (we speculate) in the MOC between warm (cold) MDV phases may be closely related to enhanced (suppressed) deep water convection in the Labrador Sea and weakened (strengthened) convection in the Greenland Sea (e.g. Dickson et al. 1996, 2002; Visbeck et al. 2002; Schlosser et al. 1991; Curry et al. 1998).
Outside the intergyre gyre region, air–ocean interactions lead to reduced (enhanced) heat fluxes from ocean to atmosphere during positive (negative) phases of MDV (e. g. Delworth and Greatbatch 2000, Fig. 5 from this study).
The well-developed atmospheric pressure centers, evident during positive phases of MDV result in intensified westerlies and trade winds (e.g. Kushnir 1994; Dickson et al. 1996; Shabbar et al. 2001; Zhang et al. 2007) and, likely, an enhanced wind-driven component of the ocean circulation.
There is extended literature devoted to mechanisms responsible for the shift from one phase of MDV to another (see, for example, in depth discussion in Marshall et al. 2001a, b). Long-term internal oceanic circulation seems to play a fundamental role in shaping climate variability on time scales from several decades to centuries. For example, our analysis demonstrates that the recent warming over the North Atlantic (0–3,000 m) is linked to both long-term (including anthropogenic) climate change and MDV, and the latter accounts for ~60% of warming in the North Atlantic since 1970. We speculate, however that the combined effect of long-term climate change and a shift to the negative, or cool, phase of MDV would result in anomalously cold North Atlantic with a corresponding climate impact in Europe.
6.3 Concluding remarks
Finally, we note that since the North Atlantic Ocean plays a crucial role in the global thermohaline circulation, multidecadal fluctuations must be taken into account when assessing long-term climate change and variability in the North Atlantic as well as over broader spatial scales. Anthropogenic climate change may be amplified or masked by multidecadal variations and these modes of variability can only be separated when the mechanisms governing them are better understood. An important caveat is that we have treated the mechanisms associated with the long term-trend and MDV separately while in reality increasing greenhouse gases likely also projects onto the multidecadal mode of variability. Advances in modeling and theory as well as continued observations are required in order to develop a deeper understanding of North Atlantic variability at multiple time scales. This will be a nontrivial task due largely to the poorly defined character of this low-frequency variability and the changing relationship with large-scale climate parameters like the NAO (Polyakova et al. 2006).
This study was supported by JAMSTEC (IP, UB, EP and XZ), NOAA/CIFAR (IP and XZ), NSF grants (IP, VA, and UB) and Stanford University (EP). We thank J. Moss for help with the graphics and K. Bryan, D. Newman, J. Walsh and J. M. Wallace for insightful comments. We really appreciate help of anonymous reviewers.
This article is distributed under the terms of the Creative Commons Attribution Noncommercial License which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited.
- Alexeev VA (2003) Sensitivity to CO2 doubling of an atmospheric GCM coupled to an oceanic mixed layer: a linear analysis. Clim Dyn 20:775–787Google Scholar
- Alexeev VA, Langen PL, Bates JR (2005) Polar amplification of surface warming on an aquaplanet in “ghost forcing” experiments without sea ice feedbacks. Clim Dyn. doi: 10.1007/s00382-005-0018-3
- Bjerknes J (1964) Atlantic air–sea interaction. Adv Geophys 20:1–82Google Scholar
- Boyer TP et al. (2006) World Ocean Database 2005. In: S. Levitus (ed) NOAA Atlas NESDIS 60. U.S. Government Printing Office, Washington, DC, 190 ppGoogle Scholar
- Divine DV, Dick C (2006) Historical variability of sea ice edge position in the Nordic Seas. J Geophys Res 111. doi: 10.1029/2004JC002851
- Intergovernmental Panel on Climate Change (2007) IPCC Fourth Assessment Report. Climate change 2007: the physical science basis. CambridgeGoogle Scholar
- Langen PL, Alexeev VA (2007) Polar amplification as a preferred response in an aquaplanet GCM. Clim Dyn. doi: 10.1007/s00382-006-0221-x
- Mann ME, Emanuel KA (2006) Atlantic hurricane trends linked to climate change. Eos 87(24): 233,238,241Google Scholar
- North GR, Bell TL, Cahalan RF, Moenig FJ (1982) Sampling errors in the estimation of empirical orthogonal functions. Mon Weather Rev 110:669–706Google Scholar
- Schott FA, Fischer J, Dengler M, Zantopp R (2006) Variability of the deep western boundary current east of the Grand Banks. Geophys Res Lett 33: L21S07. doi: 10.1029/2006GL026563
- Timmermann A, Latif M, Voss R, Grotzner A (1998) Northern Hemispheric interdecadal variability: a coupled air-sea mode. J Clim 11:1906–1931Google Scholar
- Visbeck M et al. (2002) The ocean’s response to North Atlantic oscillation variability. In: J Hurrell, Y Kushnir, G Ottersen, M Visbeck (eds) The North Atlantic oscillation: climate significance and environmental impact. Geophysical monograph series, vol 134. AGU, Washington, pp 113–146Google Scholar
- Wilks DS (2005) Statistical methods in the atmospheric sciences, 2nd edn. Academic Press, Dublin, 648 pGoogle Scholar